Dan Romero on Warpcast

Content pfp

0 reply

0 recast

0 reaction

Dan Romero pfp

People don’t want to see uninteresting replies and casts. Best way to solve for that is weight historical inbound engagement for a user, i.e. PageRank-like score for every user. Disproportionally hurts new accounts. Making it harder to get started.

52 replies

12 recasts

134 reactions

Jonny Mack pfp

maybe this could be counterbalanced by an algo boost for new accounts with interesting casts?

3 replies

2 recasts

12 reactions

Dan Romero pfp

How do you know what new casts are interesting before they have engagement?

11 replies

1 recast

6 reactions

Kyle McCollom pfp

Have to test them out to see. Just like TikTok and Reels show you a new vid every ~10 videos to feed the algo. Probably harder to measure with casts/tweets though - TikTok’s full screen takeover has its advantage.

1 reply

0 recast

0 reaction

Les Greys pfp

maybe, ask an llm?

1 reply

0 recast

2 reactions

sahil pfp

this is a real challenge. but we can try to capture upcoming users/casts based on a low threshold of engagement from highly ranked accounts 'Discovery' is a problem across all clients. an experiment-> create a 'Discovery feed' for upcoming/new users to give them visibility/chance to get initial engagement. cc @woj.eth

0 reply

0 recast

0 reaction

Jonny Mack pfp

1) display top-level casts and replies to a subset* of power badge holders 2) if no engagement, display to even fewer pb holders until eventually hiding completely 3) if engagement, display to more pb holders until eventually giving an algo boost to highly engaging new comers *could be opt-in & include warps bonus

2 replies

0 recast

4 reactions

Lazydev 🔵 XD 🎩 pfp

Lazydev 🔵 XD 🎩

Maybe they get a grace period or for like first 20 casts/replies? Uninteresting content can be detected with that sample size I think 🤔

2 replies

0 recast

1 reaction

Brent Fitzgerald pfp

Brent Fitzgerald

Maybe that’s what you’re paying for when you first sign up: the benefit of the doubt.

0 reply

0 recast

0 reaction

grant 🌈 🎩 🐸 pfp

grant 🌈 🎩 🐸

What your frens say and what they upvote

0 reply

0 recast

0 reaction

Jarrett pfp

People suggesting llm “ai” aren’t really taking into account that the types on content that people consider “interesting” change constantly 😅

0 reply

0 recast

0 reaction

TTBoyCrypto pfp

0 reply

0 recast

0 reaction

helladj™ pfp

0 reply

0 recast

0 reaction

Felipe Argento pfp

If you had infinite compute you could train an LLM on existing casts to estimate how interesting a new cast would be. For accounts younger than x days, thats the score you'd use. After that, you're back with the usual engagement metrics 😛 (but you prob don't have infinite compute, so its a very hard problem)

0 reply

0 recast

0 reaction